D M Dabrowska | PubMed Health

PubMed2000

Confidence bands for comparison of transition probabilities in a Markov chain model.

We consider graphs, confidence procedures and tests that can be used to compare transition probabilities in a Markov chain model with intensities specified by a Cox proportional hazard model. Under assumptions of this model, the regression coefficients provide information about the relative risks of covariates in one-step transitions, however, they cannot in general be used to to assess whether or not the covariates have a beneficial or detrimental effect on the endpoint events. To alleviate this problem, we consider graphical tests based on confidence procedures for a generalized Q-Q plot and for the difference between transition probabilities. The procedures are illustrated using data of the International Bone Marrow Transplant Registry.

Bone Marrow Transplantation↗

PubMed1992

Methods for comparing cumulative hazard functions in a semi-proportional hazard model.

Graphical methods based on the analysis of differences between log cumulative hazard functions are considered for a two-group semi-proportional hazard model which allows for interaction between treatments and covariates. Confidence procedures and test statistics that can be used to test for interaction and for main effects are developed. Their use is illustrated by applying them to the analysis of kidney transplant data from the University of California, San Francisco.

Azathioprine↗

PubMedDate not supplied

Identifying predictive factors in melanoma progression.

Identification of risk factors is a fundamental goal of melanoma studies. The current understanding of melanoma progression is based primarily on two-stage modeling. A multistate Markov chain process combined with Cox proportional hazard regression is used to model the melanoma progression. The model is applied to 3,434 patients initially diagnosed as AJCC stage I or stage II. Parameter estimates are obtained using Cox regression and supplemented by plots of survival probabilities. Age is associated with increased risk of progression from stage I, II to stage III and from stage III to stage IV. Males experienced an increased risk of stage I, II to stage III progression. Primary tumor located on extremities decreased the risk of all transitions. Clark's level of invasion >III and Breslow's depth >1 mm increased the risk of progression from stage I, II to stage III and stage IV. The following interactions among the prognostic factors were identified for the first time in this research: interaction of age and gender in progression from stage I, II to stage III; interactions of level and depth and site by gender were found in the progression from stage I, II to stage III; interaction of site and gender in progression from stage III to stage IV and stage IV to death. Also we identified primary site as a new prognostic factor for the progression from stage IV to death. The study employed a multistate model in order to identify prognostic factors relevant for disease progression. The unique feature is the modeling of interactions among the prognostic factors and their identification.

Adult↗